The AI Revolution in Medicine: GPT-4 and Beyond

Description:

AI is about to transform medicine. Here’s what you need to know right now.

”The development of AI is as fundamental as the creation of the personal computer. It will change the way people work, learn, and communicate–and transform healthcare. But it must be managed carefully to ensure its benefits outweigh the risks. I’m encouraged to see this early exploration of the opportunities and responsibilities of AI in medicine.”

–Bill Gates

Just months ago, millions of people were stunned by ChatGPT’s amazing abilities — and its bizarre hallucinations. But that was 2022. GPT-4 is now here: smarter, more accurate, with deeper technical knowledge. GPT-4 and its competitors and followers are on the verge of transforming medicine. But with lives on the line, you need to understand these technologies — stat.

What can they do? What can’t they do — yet? What shouldn’t they ever do? To decide, experience the cutting edge for yourself. Join three insiders who’ve had months of early access to GPT-4 as they reveal its momentous potential — to improve diagnoses, summarize patient visits, streamline processes, accelerate research, and much more. You’ll see real GPT-4 dialogues — unrehearsed and unfiltered, brilliant and blundering alike — all annotated with invaluable context, candid commentary, real risk insights, and up-to-the-minute takeaways.

• Preview a day in the life of a doctor with a true AI assistant.
• See how AI can enhance doctor-patient encounters at the bedside and beyond.
• Learn how modern AI works, why it can fail, and how it can be tested to earn trust.
• Empower patients: improve access and equity, fill gaps in care, and support behavior change.
• Ask better questions and get better answers with “prompt engineering.”
• Leverage AI to cut waste, uncover fraud, streamline reimbursement, and lower costs.
• Optimize clinical trials and accelerate cures with AI as a research collaborator.
• Find the right guardrails and gain crucial insights for regulators and policymakers.
• Sketch possible futures: What dreams may come next?

There has never been technology like this. Whether you’re a physician, patient, healthcare leader, payer, policymaker, or investor, AI will profoundly impact you — and it might make the difference between life or death. Be informed, be ready, and take charge — with this book.

See more medical ebooks at here:

Quantum Care: A Deep Dive into AI for Health Delivery and Research

Artificial Intelligence in the Genetic Diagnosis of Rare Disease

Mastering the Art of Patient Care

Artificial Intelligence for Disease Diagnosis and Prognosis in Smart Healthcare

The Thinking Healthcare System

Digital Health

Trends of Artificial Intelligence and Big Data for E-Health

Prologue

What follows is pure fiction, but everything it describes is well within the documented current capabilities of OpenAI’s GPT-4 system.

Suddenly, the patient was crashing. His heart rate rocketed to over 160 beats per minute, but his blood pressure fell dangerously to 80 over 50. Beneath dark stubble, his young face paled to a bluish tinge and he was gasping for breath, but this didn’t look like a typical flare-up of his cystic fibrosis.

Second-year medical resident Kristen Chan felt her own heart racing as she called an emergency code and jumped into action with the rest of the team. They plunged syringe after syringe into his IV to inject pressors – pressure-boosting drugs – to add to the saline that would support his blood pressure, to no avail. A drug to increase the contractile force of his heart didn’t help either.

Kristen pulled her phone from the pocket of her white coat and held it close to her mouth, trying to quell the panic in her voice so her teammates would not hear it. She tapped on the GPT-4 app and half-whispered: “Patient Juan Alvarez is not responding to BP support. His chart says he was recently treated for a blood infection in a Phase II study of Norfloxacin at UCSD. I don’t know what is happening and what to do.”

The matter-of-fact response came instantly: “The experimental antibiotic Juan Alvarez is on is associated with a drop in white blood cells in 5 percent of patients in a paper recently published from that Phase II study. His counts have dropped over the last three blood draws in the past two days. The article reports that G-CSF infusion was highly effective in reversing the drop in white blood cells. That is a tactic that could be considered.”

Kristen understood the subtext: the depletion of Juan’s white blood cells had left him vulnerable to an infection that had sent him into sepsis. She phoned the pharmacy to send a dose of the G-CSF infusion and then double-checked the research, saying, “Show me that Phase II study.”

Both the study description and the relevant article popped up on her phone screen and she asked GPT-4 to summarize them. Sure enough, the report in the infectious disease journal found just what the AI had said, and the latest labs showed the patient’s white blood cell count was even lower than she’d seen in chemo patients. Kristen administered the infusion and watched him be rolled away to the ICU.

“Gave him the G-CSF. I hope it works,” she muttered.

The response: “It is very stressful when a patient’s condition deteriorates so rapidly. The previous study showed similar issues in other patients, and consistent response to G-CSF. You are doing your best and your team is with you.”

The voice always sounded calm, avuncular. She knew artificial intelligence tools like GPT-4 could not be considered sentient, but she somehow felt like a benevolent mentor-servant with access to nearly all the world’s medical knowledge was holding her hand. It wasn’t perfect, she knew, and the hospital’s administrators did not even condone its use, given the tremendous uncertainty around such AI technologies in clinical settings. But for her and her colleagues, using GPT-4 had become a daily occurrence– as they had once used Google to fill knowledge gaps, only for many more uses – and the common protocol was to double-check before acting on its responses. GPT-4 made her feel…augmented. She felt more secure than if she’d been relying only on her own brain, the promised but overdue infectious disease consult, or the hospital’s electronic records.

“Juan will need to be moved to a different antibiotic, an even more expensive one,” she said into her phone. “I’ll need to request prior auth from his insurer. Please write the justification text for me to insert into the form.”

“Certainly.” Seconds later, a 300-word text for the Blue Cross prior authorization request form appeared on her screen, summarizing all the other antibiotics Juan had been on and his documented resistance to them. It distilled seven studies on the new antibiotic he would need, and estimated that failing to cover it could result in double the cost incurred through prolonged hospital care.

“Please send to my inbox, along with a pointer to the prior auth form,” Kristen affirmed as she walked away. “Moving on to room 65.”

“My next patient is Daria Frolova. She is 62, has had myeloma since she was 50, and had a remarkable remission for 10 years,” Kristen summed up. “Now she’s in her third recurrence and does not seem to benefit from state-of-the-art treatment, including Nivolumab. What are the options for next steps?”

“You could consider enrolling her in a new protocol for Cetuximab at the hospital’s affiliated cancer center. Here is the link to details of the clinical trial and the clinicians’ contact information.”

“Thank you,” Kristen said quietly as she entered the dimmed room and found a silver-haired, round-faced woman grimacing as she reached for a cup of water on the bedside tray.

“Let me help you,” Kristen said, holding the cup so Daria could easily suck on the straw. “How are you feeling?”

The patient swallowed two small gulps of water. “The pain comes and goes but the tired feeling never leaves,” she said.

Kristen nodded, meeting the patient’s eyes with compassion in her own. “There’s a clinical trial we think might be an option.”

“Do tell!” came a voice from behind her. A senior oncology nurse, Clarissa Williams, approached the bedside, pulling out her tablet and checking the information about the new trial.

“Mmmm hmmm,” she hummed, “Could be a fit.” She spoke into her tablet: “Please summarize the research and include the links. If it all looks good, I’ll contact the study coordinator today. But also, please include any other trials Daria should consider.”

“Certainly,” came the response. “Among 30 patients with genetically similar melanoma, so far eight are reporting remissions and seven are seeing partial remissions. Side effects generally mild but one serious hemorrhage.”

Clarissa squeezed Daria’s hand. “Fingers crossed,” she said.

“Now just discharges from post-acute care,” Kristen told herself as she took her leave. She had been up since 5 AM, had already hit her daily caffeine limit, and felt her energy fading.

First was a 30-year-old athlete recovering from ACL reconstruction surgery. As she approached his room, she heard the soft ping on her phone. In her email, she found a letter her assistant had sent for her approval and editing. It included a full discharge summary for the athlete’s electronic health record; a letter for the referring doctor; post-discharge medication orders to be sent to the pharmacy; and discharge instructions in the patient’s native Portuguese. Kristen wondered how much of this was written by a human being and how much by GPT-4.

Good. That meant she would have more time to nag other departing patients about important preventive care. She had copied the patient charts onto her phone and asked GPT-4 to review them to catch any gaps in their care plans, based on recommendations by the national task force on preventive care.

Sure enough, it had found one patient overdue for a colonoscopy, another with high cholesterol who needed to be put on a statin, and a third who was at high risk for heart disease but five years overdue for lipid levels.

Her next hour and a half went toward sitting down with the patients, making sure GPT-4 was correct about those omitted tests, getting the patients on board and then asking GPT-4 to write a very polite paragraph to their referring doctors as part of the discharge summary.

And now — now for a little “me” time.

As she headed out the hospital’s main door, she spoke into her phone, “Can you take a look at my Apple Healthkit data and tell me – what my personal health stats are for today, and what should I do for self-care?”

Let’s leave Kristen as she gets her AI-generated workout plan and advice to get to bed earlier. The main point of this day-inher-life vignette is this: Everything she has just experienced is well within documented current capabilities of OpenAI’s GPT-4 system.

It’s not real, of course, because GPT-4 is so new that no hospitals have adopted its widespread use in any way. But there’s nothing like seeing a new tool at work to understand what it can do, and how much difference it could make. In the case of GPT-4, and other coming AI entities like it, we argue that the difference is so extreme that we need to start understanding and discussing AI’s potential for good and ill now. Or rather, yesterday.

We hope you’ll come away from this book persuaded of three points:

1) GPT-4 has game-changing potential to improve medicine and health.

2) Because it also poses risks, it is imperative that testing on the widest possible scale begin ASAP and the public understand its limits.

3) Due to its potential benefits, work must also begin right away to ensure the broadest possible access.

Table of Contents
Authors’ Note

Foreword

Prologue

Chapter 1: First Contact

Chapter 2: Medicina ex Machina

Chapter 3: The Big Question: Does It “Understand?”

Chapter 4: Trust but Verify

Chapter 5: The AI-Augmented Patient

Chapter 6: So Much More: Math, Coding, and Logic

Chapter 7: The Ultimate Paperwork Shredder

Chapter 8: Smarter Science

Chapter 9: Safety First

Chapter 10: The Big Black Bag

Epilogue